Abstract
The cervical vertebrae play a critical role in the structural and functional integrity of the human body. Their morphology is closely linked to spinal pathologies, injuries, and developmental conditions. This study investigates the classification of cervical vertebral maturation (CVM) stages using three convolutional neural networks (CNNs): ShuffleNet, AlexNet, and DenseNet. A dataset of 1018 cephalometric radiographs from patients aged 6 to 56 years was analyzed and categorized into three age-based stages. Each image was annotated to ensure accurate labeling of C2, C3, and C4 vertebrae. Features extracted by the CNNs were further processed using Principal Component Analysis (PCA) and classified with Support Vector Machines (SVM). The highest performance was achieved with ShuffleNet features combined with PCA and SVM, reaching an accuracy of 72.69%.